In recent years, methods for measuring diet quality have evolved and a number of scoring systems or indices to this effect have emerged. This relatively new concept involves the assessment of both quality and variety of the entire diet, enabling examination of associations between whole foods and health status, rather than just nutrients. Diet quality is measured by scoring food patterns in terms of how closely they align with national dietary guidelines and how diverse the variety of healthy choices is within core food groups or equivalent international groupings. More refined scoring methods allow both protective dietary patterns and unfavourable intakes to be identified. As diet quality and variety scores have been examined in association with health outcomes cross-sectionally and to predict such outcomes longitudinally, nutrition interventions could potentially be developed to target improvements in the most critical aspects of an individual’s or population’s food intake; for example, targeting an increase in fruit and vegetable consumption if an index shows a strong relationship between low intake and CVD.
Kant has previously published two reviews of diet quality indices(Reference Kant1, Reference Kant2), indicating a need for additional validation studies to assess the effectiveness of these indices in predicting nutritional and health status. Thus the aims of the present review were to:
1. Describe current diet quality tools and their applications.
2. Analyse whether higher diet quality scores are associated with lower levels of morbidity and mortality.
3. Examine how robust the studies of association between diet quality and morbidity and mortality are in relation to their findings.
Methods
Search strategy for identification of studies
A systematic review of published English-language literature from 2004 to 2007 was conducted. Relevant literature prior to this time was identified from the reviews previously published by Kant(Reference Kant1, Reference Kant2).
The following electronic databases were searched: MEDLINE, Cochrane, EMBASE (Excerpta Medica Database), CINAHL (Cumulative Index to Nursing and Allied Health Literature) and ProQuest. Scopus was used to identify studies that had cited Kant’s 2004 review(Reference Kant2). The MeSH (Medical Subject Headings of the National Library of Medicine) keyword search terms included diet, dietary, quality, variety, diversity, pattern, score, indicator, index, guideline, Healthy Eating Index, Alternative Healthy Eating Index, Recommended Food Score, chronic disease, cancer, cardiovascular disease; and variations of these.
Selection of studies
The current review used specific selection criteria for retrieval of studies. Inclusion criteria were: (i) studies of a longitudinal/cohort, case–control or cross-sectional nature; (ii) human subjects; (iii) adults; (iv) English language; (v) literature published from 2004 onwards; and (vi) use of theoretically defined dietary patterns (food indices/scores) or a measure of diet quality created a priori and based on current nutrition knowledge(Reference Waijers, Feskens and Ocke3). Pertinent literature published in one of Kant’s previous reviews(Reference Kant2) was adapted and cited to facilitate the aim of the current review.
Specific exclusion criteria included: (i) studies of non-English language; (ii) studies conducted with children or pregnant women; (iii) intervention studies; (iv) published prior to 2004, with the exception of Kant(Reference Kant1, Reference Kant2); (v) studies on animals; and (vi) studies that primarily use dietary patterns derived a posteriori from food consumption data, based and defined empirically or statistically, such as with cluster analysis and factor analysis(Reference Waijers, Feskens and Ocke3).
Articles were retrieved from the electronic search if information contained in the title, descriptor/MeSH headings and abstract appeared to be consistent with the inclusion criteria. The methodological quality of the articles was critically appraised and a summary was tabulated based on JBI (Joanna Briggs Institute) observational critical appraisal criteria of a random sample, clear inclusion criteria, objective assessment of outcomes, sufficient description of group comparison and appropriate statistical analyses; while study details, key findings and risk reduction ratios were extracted(4, 5) and tabulated.
Results
After retrieval of articles, nine articles were selected. These studies were combined with nineteen studies, previously reviewed by Kant(Reference Kant1, Reference Kant2), giving a total of twenty-eight studies included. Table 1 describes the diet quality indices or tools used. Table 2 summarises the quality assessment. Tables 3 and 4 summarise the main features of studies validating the indices cross-sectionally or in a case–control design with biomarker or health outcomes, including the dietary assessment measure, population, main results and study limitations.
Table 1 Description of diet quality indices
Table 2 Quality assessment of included studies
U, unclear; N, no; Y, yes; CPH, Cox proportional hazard model.
*All studies Yes for statistical tests.
Table 3 Cross-sectional validation studies of diet quality scores with biomarkers and biological outcomes
NHANES III, Third National Health and Nutrition Survey; MMSE, Mini-Mental State Examination; RBC, red blood cell; HDL-C, HDL cholesterol; LDL-C, LDL cholesterol.
Diet quality indices and scores
A total of twenty-five indices of overall diet quality and/or variety were identified (Table 1). The major indices include the Healthy Eating Index (HEI)(Reference Hann, Rock, King and Drewnowski6), the Healthy Diet Indicator (HDI)(Reference Huijbregts, Feskens, Rasanen, Fidanza, Nissinen, Menotti and Kromhout7), the Healthy Food Index (HFI)(Reference Osler, Heitmann, Gerdes, Jorgensen and Schroll8), the Recommended Food Score (RFS)(Reference Kant, Schatzkin, Graubard and Schairer9), the Diet Quality Index (DQI)(Reference Seymour, Calle, Flagg, Coates, Ford and Thun10), the Diet Quality Score (DQS)(Reference Fitzgerald, Dewar and Veugelers11) and the Mediterranean Diet Score (MDS)(Reference Trichopoulou, Kouris-Blazos, Wahlqvist, Gnardellis, Lagiou, Polychronopoulos, Vassilakou, Lipworth and Trichopoulos12). Table 1 describes each index, the variations or modifications derived from them and validation studies(Reference Hann, Rock, King and Drewnowski6, Reference Weinstein, Vogt and Gerrior13–Reference Toft, Kristoffersen, Lau, Borch-Johnsen and Jorgensen15). The Diet Quality Index International (DQI-I)(Reference Kim, Haines, Seiga-Riz and Popkin16) was not included because rather than examine associations with disease outcomes it compared diet quality between countries.
The majority of indices are based on national nutrition recommendations and national dietary guidelines specific to the country where the tool was developed. Adherence to these recommendations is assessed by the diet quality/variety score and then commonly compared with nutrient intakes (not reported here) and the risk for various health outcomes, including biomarkers of disease, mortality and chronic diseases such as CVD and cancer (Table 2). While most indices have been created for use with the US population, indices based on the Mediterranean diet have emerged due to recent research highlighting associations with reduced risk of CVD and some types of cancer(Reference Waijers, Feskens and Ocke3, Reference Trichopoulou, Kouris-Blazos, Wahlqvist, Gnardellis, Lagiou, Polychronopoulos, Vassilakou, Lipworth and Trichopoulos12, Reference Trichopoulou, Costacou, Bamia and Trichopoulos17–Reference Knoops, Groot de, Fidanza, Alberti-Fidanza, Kromhout and van Staveren20).
Kant highlighted in 1996(Reference Kant1) that the construction of diet quality indices has taken three major approaches: (i) based on food groups or specific foods; (ii) based on nutrient intakes; or (iii) derived from combinations of foods and nutrient intakes. In 2007 Waijers et al. added that current scores were based on adherence to established national dietary guidelines or a Mediterranean pattern(Reference Waijers, Feskens and Ocke3). Depending on inclusion or exclusion of specific foods and/or nutrients, diet quality indices can be generated to reflect a dietary intake that is healthy, unhealthy or a combination of both(Reference Waijers, Feskens and Ocke3). Most indices, including the HEI and DQI, are based on both food groups and nutrients, while some, such as the HFI, are based on foods and food groups (Table 1). The food items and groups selected as components of the indices include vegetables, fruits, cereals and grains, meat products and dairy products, with some specifically including fish and olive oil(Reference Waijers, Feskens and Ocke3). Common nutrient components used in scores include total fat, ratios of fat types (ratio of saturated fat to mono- or polyunsaturated fat) and cholesterol(Reference Waijers, Feskens and Ocke3). Alcohol is used in some indices(Reference Trichopoulou, Kouris-Blazos, Wahlqvist, Gnardellis, Lagiou, Polychronopoulos, Vassilakou, Lipworth and Trichopoulos12, Reference Trichopoulou, Costacou, Bamia and Trichopoulos17–Reference McCullough, Feskanich, Stampfer, Giovannucci, Rimm, Hu, Spiegelman, Hunter, Colditz and Willett22), predominantly Mediterranean indices, in recognition of the proposed beneficial effects and to evaluate associations with health outcomes(Reference Waijers, Feskens and Ocke3). Other nutrients used include Na, dietary fibre, protein and complex carbohydrates(Reference Waijers, Feskens and Ocke3).
Dietary variety or diversity is included in some indices (HEI, HEI-f, DQI-R, total and specific food group diversity and its variations), in addition to foods and nutrients, with higher scores awarded for a more varied diet(Reference Waijers, Feskens and Ocke3). It has been proposed that variety within food groups may be considered a better indicator of more healthful outcomes(Reference Hann, Rock, King and Drewnowski6, Reference Weinstein, Vogt and Gerrior13, Reference Fung, Hu, McCullough, Newby, Willett and Holmes21, Reference McCullough, Feskanich, Rimm, Giovannucci, Ascherio, Variyan, Spiegelman, Stampfer and Willett23–Reference Slattery, Berry, Potter and Caan27); however, this is not considered important by some authors(Reference Waijers, Feskens and Ocke3).
Scoring methods and cut-off points to assess diet quality vary across indices. The various MDS indices use sex-specific median cut-off points(Reference Trichopoulou, Kouris-Blazos, Wahlqvist, Gnardellis, Lagiou, Polychronopoulos, Vassilakou, Lipworth and Trichopoulos12, Reference Trichopoulou, Costacou, Bamia and Trichopoulos17–Reference Fung, Hu, McCullough, Newby, Willett and Holmes21). Median cut-off points allow subjects to be scored positively or negatively for each item(Reference Waijers, Feskens and Ocke3). Indices such as the Dietary Guidelines Index (DGI)(Reference Harnack, Nicodemus, Jacobs and Folsom28) use lower, intermediate and upper cut-off levels(Reference Waijers, Feskens and Ocke3), while the HDI uses dichotomous values to determine whether dietary recommendations have been met or not. For example, with the HDI, if a person consumed 27–40 g dietary fibre daily, he/she was awarded 1 point for the recommendation being met(Reference Waijers, Feskens and Ocke3). It is debatable whether cut-off boundaries, dichotomous values or continuous variables allow for a better evaluation of adherence to recommendations or examination of associations with disease outcomes.
Health outcomes associated with diet quality indices
Diet quality indices can also be used to measure risk of various health outcomes, including biomarkers of disease (Table 3), mortality and chronic diseases (Tables 4 and 5). These are discussed by major dietary index type to allow better comparison of the study results.
Table 4 Case–control studies comparing diet quality with morbidity and mortality
Table 5 Prospective cohort studies with mortality and morbidity as outcome, ordered by increasing length of follow-up
EPIC, European Prospective Investigation into Cancer and Nutrition; NHANES I; First National Health and Nutrition Survey; HALE, Health Ageing: a Longitudinal study in Europe; PA, physical activity; ER, oestrogen receptor.
Healthy Eating Index
The HEI was used in three studies(Reference Hann, Rock, King and Drewnowski6, Reference Weinstein, Vogt and Gerrior13, Reference Fung, Hu, McCullough, Newby, Willett and Holmes21) and was associated with biomarkers such as carotenoids (except lycopene), cholesterol and vitamins C and E(Reference Hann, Rock, King and Drewnowski6, Reference Weinstein, Vogt and Gerrior13). The strongest associations occurred for the biomarkers of vegetable and fruit consumption(Reference Hann, Rock, King and Drewnowski6, Reference Weinstein, Vogt and Gerrior13, Reference Fung, Hu, McCullough, Newby, Willett and Holmes21). Mean HEI was significantly higher in the 42 % of participants who consumed dietary supplements(Reference Weinstein, Vogt and Gerrior13). In one study, the HEI was used among various other indices (AHEI, DQI-R, RFS and aMED) to measure incidence of breast cancer(Reference Fung, Hu, McCullough, Newby, Willett and Holmes21). However, it was shown to have limited ability to predict risk for oestrogen receptor (ER)-negative breast cancer(Reference Fung, Hu, McCullough, Newby, Willett and Holmes21).
An Alternative HEI (AHEI) was used in two studies, with varied associations with cancer incidence(Reference Fung, Hu, McCullough, Newby, Willett and Holmes21, Reference McCullough, Feskanich, Stampfer, Giovannucci, Rimm, Hu, Spiegelman, Hunter, Colditz and Willett22). It did not predict all-site cancer risk in a study of US adults(Reference McCullough, Feskanich, Stampfer, Giovannucci, Rimm, Hu, Spiegelman, Hunter, Colditz and Willett22) but was inversely associated with chronic disease risk in men and women (39 % and 28 % reduced CVD risk in the highest AHEI quintile for men and women, respectively, with a weaker relationship in women). AHEI was associated with lower risk of ER-negative breast cancer in a study of postmenopausal women, with an 11 % reduction in risk for each 10 % increase in score(Reference Fung, Hu, McCullough, Newby, Willett and Holmes21).
HEI derived from an FFQ (HEI-f) was used in two studies(Reference McCullough, Feskanich, Rimm, Giovannucci, Ascherio, Variyan, Spiegelman, Stampfer and Willett23, Reference McCullough, Feskanich, Stampfer, Rosner, Hu, Hunter, Variyan, Colditz and Willett24). HEI-f was associated with an 11 % reduction in major chronic disease risk in men, but not women. These studies also found a reduced CVD risk in men (28 %) and women (14 %), but there were no associations with reduced cancer risk(Reference McCullough, Feskanich, Rimm, Giovannucci, Ascherio, Variyan, Spiegelman, Stampfer and Willett23, Reference McCullough, Feskanich, Stampfer, Rosner, Hu, Hunter, Variyan, Colditz and Willett24).
Healthy Diet Indicator
The HDI was used in three studies(Reference Huijbregts, Feskens, Rasanen, Fidanza, Nissinen, Menotti and Kromhout7, Reference Knoops, Groot de, Fidanza, Alberti-Fidanza, Kromhout and van Staveren20, Reference Huijbregts, Feskens, Rasanen, Fidanza, Alberti-Fidanza, Nissinen and Giampaoli29). One study used cognitive function as the outcome measure and found in four of five cohorts that a higher HDI was associated with lower cognitive impairment. However, this was statistically significant in only one cohort(Reference Huijbregts, Feskens, Rasanen, Fidanza, Alberti-Fidanza, Nissinen and Giampaoli29). Two studies found the HDI to be inversely associated with all-cause mortality(Reference Huijbregts, Feskens, Rasanen, Fidanza, Nissinen, Menotti and Kromhout7, Reference Knoops, Groot de, Fidanza, Alberti-Fidanza, Kromhout and van Staveren20). One found a 13 % reduction in all-cause mortality as well as an 18 % risk reduction in CVD mortality, with no reduction in cancer mortality risk(Reference Huijbregts, Feskens, Rasanen, Fidanza, Nissinen, Menotti and Kromhout7). The other found that the HDI had a significant inverse relationship with all-cause mortality after a 10-year follow-up(Reference Knoops, Groot de, Fidanza, Alberti-Fidanza, Kromhout and van Staveren20).
Healthy Food Index
The HFI was used in two assessments of the same population, but reported associations with different outcomes(Reference Osler, Heitmann, Gerdes, Jorgensen and Schroll8, Reference Osler, Andreasen, Heitmann, Hiodrup, Gerdes, Jorgensen and Schroll30). The first measured CVD mortality(Reference Osler, Heitmann, Gerdes, Jorgensen and Schroll8) while the second measured risk of CHD(Reference Osler, Andreasen, Heitmann, Hiodrup, Gerdes, Jorgensen and Schroll30). HFI was inversely associated with all-cause mortality in men and women; however, this was attenuated after adjustment for smoking, BMI, physical activity, alcohol intake and education level(Reference Osler, Heitmann, Gerdes, Jorgensen and Schroll8). It was not related to cause-specific mortality(Reference Osler, Heitmann, Gerdes, Jorgensen and Schroll8). The second report found an insignificant, inverse relationship with CHD risk(Reference Osler, Andreasen, Heitmann, Hiodrup, Gerdes, Jorgensen and Schroll30).
The Healthy Food and Nutrient Index (HFNI) was used in one study(Reference Bazelmans, De Henauw, Matthys, Dramaix, Kornitzer, De Backer and Leveque31). It had a significant association with mortality in men after adjustment for all-cause mortality risk factors. This index showed no significant associations in women.
Recommended Food Score
The original RFS was used in three studies(Reference Kant, Schatzkin, Graubard and Schairer9, Reference McCullough, Feskanich, Stampfer, Giovannucci, Rimm, Hu, Spiegelman, Hunter, Colditz and Willett22, Reference Mai, Kant, Flood, Lacey, Schairer and Schatzkin32). RFS was inversely associated with all-cause mortality and cause-specific mortality in a study of US women(Reference Kant, Schatzkin, Graubard and Schairer9). There was a 30 % reduced risk of all-cause and cause-specific mortality in the highest quartile, after age and multivariate adjustment(Reference Kant, Schatzkin, Graubard and Schairer9). One of the studies that measured AHEI also found that the RFS was inversely associated with chronic disease risk in men, with a 23 % lower CVD risk, and that this was a smaller association than with AHEI(Reference McCullough, Feskanich, Stampfer, Giovannucci, Rimm, Hu, Spiegelman, Hunter, Colditz and Willett22). The RFS was not related to major chronic disease outcomes in women and did not predict cancer in either gender(Reference McCullough, Feskanich, Stampfer, Giovannucci, Rimm, Hu, Spiegelman, Hunter, Colditz and Willett22). In contrast to the previous study, another in US women found the RFS to be inversely associated with total mortality, cancer mortality, and mortality from colorectal, lung and breast cancer, with the highest RFS quartile having a reduced incidence of lung cancer(Reference Mai, Kant, Flood, Lacey, Schairer and Schatzkin32).
Two studies using a slightly varied version of RFS also found associations with mortality and cancer(Reference Fung, Hu, McCullough, Newby, Willett and Holmes21, Reference Michels and Wolk33). It was inversely associated with all-cause and cause-specific mortality, particularly CHD and stroke, with 42 % reduced all-cause mortality(Reference Michels and Wolk33). One study previously mentioned found the RFS to be associated with a lower risk of ER-negative breast cancer (12 % reduction for each 10 % increase in score), and that it had the strongest association among all the indices used (HEI, AHEI, DQI-R and aMED)(Reference Fung, Hu, McCullough, Newby, Willett and Holmes21).
A Not Recommended Food Score (NRFS) was used in a previous study(Reference Michels and Wolk33) using the RFS, which showed no relationships with all-cause mortality, CHD and stroke mortality.
Diet Quality Index
The original DQI was used in one study of US adults(Reference Seymour, Calle, Flagg, Coates, Ford and Thun10). After adjustment for multiple covariates the DQI was associated only with circulatory disease mortality in women and it was unrelated to cancer mortality. An improved index, the DQI Revised (DQI-R), was limited in predicting risk of breast cancer(Reference Fung, Hu, McCullough, Newby, Willett and Holmes21). Another version of the DQI was modified for use in a Mediterranean population with dietary biomarkers as an outcome(Reference Gerber, Scali, Michaud, Durand, Astre, Dallongeville and Romon14). It was found to be significantly correlated with EPA and DHA only. This association was still apparent after adjusting for tobacco use, and a correlation with β-carotene became significant after this adjustment.
Diet Quality Score
Two variations of the DQS have been used. The first measured all-sites cancer incidence over 8·3 years and found a significant inverse relationship, estimating that the incidence of cancer could be reduced by ∼35 % if diet quality was improved(Reference Fitzgerald, Dewar and Veugelers11). The second, using CVD biomarkers, found DQS to be negatively associated with total cholesterol, LDL cholesterol, TAG, homocysteine and absolute risk of IHD after adjustment for age, sex, physical activity level and smoking(Reference Toft, Kristoffersen, Lau, Borch-Johnsen and Jorgensen15). DQS was positively associated with HDL cholesterol concentrations. Another score, the Dietary Quality (DQ), was used to examine associations with survival after age 45 years(Reference Murphy, Davis, Neuhaus and Lein34). The latter study in US adults found that DQ was inversely associated with all-cause mortality, with the relative hazards of death for men aged 55–64 years with the worst DQ after 10-year follow-up being 1·9 compared with those with better diets. The relative hazards of death for men aged 45–54 and 65–74 years were 1·5 and 1·4, respectively(Reference Murphy, Davis, Neuhaus and Lein34).
Dietary Guidelines Index
The DGI was used in one study to measure all-site cancer and site-specific cancer incidence and mortality(Reference Harnack, Nicodemus, Jacobs and Folsom28). Higher DGI was associated with a significantly lower incidence of all-site cancer and site-specific cancers. This association was no longer significant once guidelines such as be physically active each day were excluded and it was analysed against only diet-based dietary guidelines and cancer incidence, with the exception of lung and bronchus.
Mediterranean Diet Score
Numerous versions of MDS have been constructed, most with slight variations from the original, mainly in the scoring of fish, legumes and vegetables, as detailed in Table 1(Reference Trichopoulou, Kouris-Blazos, Wahlqvist, Gnardellis, Lagiou, Polychronopoulos, Vassilakou, Lipworth and Trichopoulos12). Six studies used the MDS with most measuring all-cause mortality and cancer mortality as outcomes(Reference Trichopoulou, Kouris-Blazos, Wahlqvist, Gnardellis, Lagiou, Polychronopoulos, Vassilakou, Lipworth and Trichopoulos12, Reference Trichopoulou, Costacou, Bamia and Trichopoulos17–Reference Knoops, Groot de, Fidanza, Alberti-Fidanza, Kromhout and van Staveren20). The first version found that every 1-unit increase in score was significantly associated with a 17 % overall mortality reduction(Reference Trichopoulou, Kouris-Blazos, Wahlqvist, Gnardellis, Lagiou, Polychronopoulos, Vassilakou, Lipworth and Trichopoulos12). A second version was used in two studies(Reference Trichopoulou, Costacou, Bamia and Trichopoulos17, Reference Lagiou, Trichopoulos, Sandin, Lagiou, Mucci, Wolk, Weiderpass and Adami18). The first study on adults in Greece found the MDS to be associated with reduced total mortality (25 % reduction for a 2-point increase in score) and inversely associated with mortality due to CHD and cancer(Reference Trichopoulou, Costacou, Bamia and Trichopoulos17). The second study in Swedish women also found an inverse relationship, with a substantial reduction in all-cause mortality (23 %) and cancer mortality (29 %) in women aged 40–49 years at enrolment(Reference Lagiou, Trichopoulos, Sandin, Lagiou, Mucci, Wolk, Weiderpass and Adami18). This association was not found in women who were aged 30–40 years at enrolment(Reference Lagiou, Trichopoulos, Sandin, Lagiou, Mucci, Wolk, Weiderpass and Adami18). A third version of the MDS was used to measure all-cause mortality as an outcome(Reference Osler and Schroll19). It too was inversely associated with all-cause mortality; a 1-unit increase in score being associated with 21 % reduced mortality. The study also found higher plasma carotene levels in subjects with higher scores and a negative association between plasma carotene and mortality(Reference Osler and Schroll19). The fourth version of the MDS was used in a study that also used the HDI and Mediterranean Adequacy Index (MAI)(Reference Knoops, Groot de, Fidanza, Alberti-Fidanza, Kromhout and van Staveren20). All three indices had a significant inverse relationship with all-cause mortality. An Alternative MDS (aMDS) was used in a previous study on breast cancer and was associated with lower risk of ER-negative breast cancer (7 % reduction for each 10 % increase in score)(Reference Fung, Hu, McCullough, Newby, Willett and Holmes21).
Mediterranean Adequacy Index
One study used the MAI to measure all-cause mortality, cancer and other causes of mortality(Reference Knoops, Groot de, Fidanza, Alberti-Fidanza, Kromhout and van Staveren20). Only all-cause mortality had a significant inverse relationship with MAI, with associations found to be stronger in Northern than in Southern Europe(Reference Knoops, Groot de, Fidanza, Alberti-Fidanza, Kromhout and van Staveren20).
Total and specific food group diversity
Three case–control studies of colon and colorectal cancer used total and specific food group diversity(Reference Fernandez, Negri, La Vecchia and Franceschi25–Reference Slattery, Berry, Potter and Caan27). The first study from Italy found total diet diversity reduced the risk of colorectal cancer by 30 % in the highest quartile of diversity(Reference Fernandez, D’Avanzo, Negri, Franceschi and La Vecchia26). Relative risk appeared to increase by 1·4 with diversity of foods in the carbohydrate group(Reference Fernandez, D’Avanzo, Negri, Franceschi and La Vecchia26). The second study found that total diet diversity had no association with colon cancer in US adults(Reference Slattery, Berry, Potter and Caan27). However, men had a 50 % increased risk of colon cancer with greater diet diversity in the meat/fish/poultry/egg group and refined grains group(Reference Slattery, Berry, Potter and Caan27). Women had a 20 % lower risk if diet diversity was high in the vegetable group(Reference Slattery, Berry, Potter and Caan27). The third study, also in Italy, found total diet diversity to be inversely related to colon cancer risk in men (35 % reduction) and that diversity within the vegetable group had an inverse relationship with both colon and rectal cancers(Reference Fernandez, Negri, La Vecchia and Franceschi25).
Discussion
The majority of studies reviewed found that diet quality scores were inversely related to health outcomes, with a protective effect of moderate magnitude. In the studies that did find associations, all-cause mortality was reduced by 17–42 %, CVD mortality by 18–53 %, CVD risk by 14–28 %, cancer mortality by 13–30 %, and all-cancer risk by 7–35 %. The predictive capacity of most indices appears to be in a similar range. However, it is difficult to directly compare results due to the variability of population groups, length of study follow-up periods, dietary measurement methods, index scoring methods and approaches to adjustment for confounders(Reference Waijers, Feskens and Ocke3). Relative risks were consistently attenuated after adjustment for factors such as BMI, physical activity, age, education, socio-economic status and smoking.
Associations between scores and morbidity and/or mortality varied between genders within studies, and were generally stronger and broader in their associations with adverse health outcomes for men. This suggests that this is a real disparity, although differential gender bias in the original dietary assessment method used to assess usual intake is a possibility.
Specific outcomes, such as cancer risk, were not predicted as strongly or consistently as all-cause mortality or CVD risk. Those indices that did predict risk came from tools based on either a US dietary pattern (RFS(Reference Mai, Kant, Flood, Lacey, Schairer and Schatzkin32), AHEI(Reference Fung, Hu, McCullough, Newby, Willett and Holmes21) and Total Diet Diversity(Reference Fernandez, Negri, La Vecchia and Franceschi25, Reference Fernandez, D’Avanzo, Negri, Franceschi and La Vecchia26)) or Mediterranean dietary patterns (MDS(Reference Lagiou, Trichopoulos, Sandin, Lagiou, Mucci, Wolk, Weiderpass and Adami18) and aMED(Reference Fung, Hu, McCullough, Newby, Willett and Holmes21)). One possible reason for the lack of association is study duration, with shorter follow-up obviously not able to capture the latency period prior to cancer detection(Reference Fitzgerald, Dewar and Veugelers11, Reference Trichopoulou, Kouris-Blazos, Wahlqvist, Gnardellis, Lagiou, Polychronopoulos, Vassilakou, Lipworth and Trichopoulos12, Reference Osler and Schroll19, Reference McCullough, Feskanich, Rimm, Giovannucci, Ascherio, Variyan, Spiegelman, Stampfer and Willett23, Reference McCullough, Feskanich, Stampfer, Rosner, Hu, Hunter, Variyan, Colditz and Willett24, Reference Bazelmans, De Henauw, Matthys, Dramaix, Kornitzer, De Backer and Leveque31). Two of the studies that did find associations with cancer risk were case–control studies that measured the diet diversity of participants after initial cancer diagnosis through administration of an FFQ covering the previous year(Reference Fernandez, Negri, La Vecchia and Franceschi25, Reference Fernandez, D’Avanzo, Negri, Franceschi and La Vecchia26). However, dietary intake in these instances is likely to be confounded by recall bias and it is difficult to ascertain with this study design in which direction the risk assessment is likely to be affected. All indices have limitations due to the methods with which dietary intake was measured. Instruments such as the 24 h recall and FFQ all have limitations including over- or under-reporting, and in some studies only a brief FFQ was originally used(Reference Osler, Heitmann, Gerdes, Jorgensen and Schroll8, Reference Fernandez, D’Avanzo, Negri, Franceschi and La Vecchia26, Reference Osler, Andreasen, Heitmann, Hiodrup, Gerdes, Jorgensen and Schroll30) or sometimes a single 24 h recall was used as the indicator of usual dietary intake(Reference Weinstein, Vogt and Gerrior13). Having an initial measurement of dietary intake at baseline in cohort studies is a more sensitive way to ascertain prospective associations with disease risk. However, this approach requires long periods of follow-up and is also not without bias and limitations. To strengthen comparisons across studies, researchers ideally could base their index of diet quality on one of the tools commonly used or use more than one of the indices in statistical analyses, as some researchers have done(Reference Michels and Wolk33). The critical appraisal of study quality found that most studies met at least three of five criteria, with about a third meeting at least four. While outcomes were usually reported objectively and statistical methods appeared appropriate, descriptions of the population sampling, inclusion criteria and subgroups were often poor.
Many components of indices such as the HEI are derived from epidemiological associations with reduced risk of CVD and its risk factors, as opposed to cancer. This may explain why indices appear to have a better risk prediction for CVD(Reference McCullough, Feskanich, Rimm, Giovannucci, Ascherio, Variyan, Spiegelman, Stampfer and Willett23). Associations may also be less strong between index components and cancer risk. Most indices have been developed to measure diet quality and adherence to national dietary guidelines (HFNI, RFS, DQS and DGI). While they attempt to show association with higher diet quality, it is difficult to predict the likely degree of association with disease risk if the index was not specifically designed for this purpose(Reference Waijers, Feskens and Ocke3). For example in the HEI, grains are not differentiated between those that are refined or unrefined, which limits the ability to show effects between whole grains and diseases such as diabetes and heart disease(Reference McCullough, Feskanich, Rimm, Giovannucci, Ascherio, Variyan, Spiegelman, Stampfer and Willett23).
Indices have their own strengths and limitations which may have affected results in the studies. Indices that have a small scoring scale, such as the DQI and HEI, appear to be less sensitive in this evaluation of indices and fail to capture extremes and intrinsic characteristics of food behaviours or eating patterns. Their discrete distributions within index scoring patterns may reduce their power as a predictor(Reference Panagiotakos, Pitsavos and Stefanadis35). Both the DQI and the HEI have been revised (DQI-R and AHEI) to improve disease risk prediction. AHEI was twice as strong as HEI in its overall chronic disease risk prediction, primarily CVD, in men and women in the USA(Reference McCullough and Willett36). When selecting or developing a dietary quality index, important considerations include: (i) defining the tool’s purpose, such as to capture a global indicator of adherence to specific aspects of a healthful diet or to measure associations with aspects of diet that increase disease risk; (ii) identifying a scoring system that weights the subscales appropriately; and (iii) measuring potential confounders to allow appropriate adjustment for confounders. Further, consideration should be given to including a validation of the score using appropriate biomarkers and disease risk factor assessment. For example, a dietary quality index to measure associations with CVD outcomes could include index components that score type and quantity of dietary fat, measure red blood cell membrane fatty acids as the dietary biomarkers, plasma lipids as the disease risk factors and confounders of CVD as previously described.
Other limitations of study design include small cohort size(Reference Hann, Rock, King and Drewnowski6, Reference Trichopoulou, Kouris-Blazos, Wahlqvist, Gnardellis, Lagiou, Polychronopoulos, Vassilakou, Lipworth and Trichopoulos12, Reference Gerber, Scali, Michaud, Durand, Astre, Dallongeville and Romon14, Reference Osler and Schroll19), cross-sectional assessments only with no follow-up(Reference Hann, Rock, King and Drewnowski6, Reference Weinstein, Vogt and Gerrior13–Reference Toft, Kristoffersen, Lau, Borch-Johnsen and Jorgensen15, Reference Huijbregts, Feskens, Rasanen, Fidanza, Alberti-Fidanza, Nissinen and Giampaoli29) and using mortality as the sole outcome, given that fatality may be confounded by early diagnosis, treatment and medical care(Reference McCullough, Feskanich, Stampfer, Rosner, Hu, Hunter, Variyan, Colditz and Willett24). Therefore indices need to be carefully interpreted to ensure limitations are acknowledged.
Conclusions
Indices of diet quality are used to measure associations with biomarkers and health outcomes. We found that lower diet quality scores are consistently associated with higher rates of all-cause mortality and selected disease-specific rates or mortality. The associations are attenuated when adjusted for common confounding variables but still remain significant, and appear to be stronger in men and for all-cause and CVD mortality. However, the limitations of these indices and the specific context in which they are used need to be considered when interpreting results and comparing studies. Future validation studies need to examine associations between nutritional biomarkers and intermediary disease risk factors. Not only will this improve the validity of the diet quality indices, but it will also increase their potential practical applications in both clinical and public health contexts.
As there are numerous diet quality indices and variations of each method, it is recommended that researchers model indices on existing tools and select more than one when testing associations with health outcomes. This will allow examination of how robust findings are across the indices and facilitate comparison across studies.
Finally, there is enough evidence to recommend that diet quality tools should be adapted for use in clinical dietetic practice and for self-evaluation of dietary intake, particularly those scored in a way that identifies which foods need to be increased to obtain a more healthful score and therefore potentially reduce chronic disease risk.
Acknowledgements
The present review was undertaken in partial requirement for the degree of Bachelor of Nutrition and Dietetics (A.W.), School of Health Sciences, Faculty of Health, University of Newcastle. This work was completed at the University of Newcastle. There was no funding for this review. There are no conflicts of interests. C.E.C. and A.W. designed the protocol; A.W. searched the literature and drafted the manuscript; C.E.C. and A.W. revised the manuscript.
